Real-World Aircraft Recognition Based on RF Fingerprinting With Few Labeled ADS-B Signals

Automatic Dependent Surveillance-Broadcast (ADS-B) is considered as a radar alternative in air transport systems, yet lack of authentication and encryption make it vulnerable to attack. In this paper, we propose a noise-robust radio frequency fingerprinting (RFF) approach for practical aircraft identification scenarios with a small sample size to address these issues. We develop a preprocessing method to improve noise-robustness by zero-padding useless signals and use a Siamese Networks-based few-shot training scheme for RFF recognition. The method is evaluated on a real-world ADS-B dataset, showing that signal preprocessing increases aircraft recognition accuracy by approximately 20% compared to using raw signals directly in small sample cases. Even with tens of labeled samples, our method achieves over 90% accuracy, outperforming other CNN identifiers by over 30%.